Machine learning-based assessment of the healthy human gut mycobiota landscape using ITS1 DNA metabarcoding data

Area of Science:

• Human microbiome research
• Mycology
• Computational biology

Background:

• The gut microbiome influences host health, with a known bidirectional relationship.
• DNA metabarcoding is key for studying microbiome imbalances (dysbiosis).
• The fungal component (mycobiome) is less understood than the prokaryotic microbiome.

Purpose of the Study:

• To comprehensively analyze the human gut mycobiome using DNA metabarcoding.
• To integrate Machine Learning (ML) and explainable Artificial Intelligence (XAI) for health status prediction.
• To identify fungal biomarkers for non-invasive health diagnostics.

Main Methods:

• Analysis of ~1,500 publicly available ITS1 sequences using DNA metabarcoding.
• Application of conventional statistical methods alongside ML and XAI.
• Development of a multiview analytical framework for mycobiome data.

Main Results:

• ML models achieved >80% accuracy in predicting host health status.
• Specific fungal genera (Eurotium, Aureobasidium, Candida, Cutaneotrichosporon) were identified as key predictors.
• A novel analytical framework was applied to public mycobiome data.

Conclusions:

• Gut fungal community profiling shows potential as a non-invasive diagnostic tool.
• ML and XAI can effectively predict host health from mycobiome data.
• This study provides a foundation for further research into the mycobiome's role in health and disease.